Post on 13-Feb-2022
KNOWLEDGE SPILLOVERS THROUGH MULTINATIONAL FIRMSIN HIGH-TECH AND LOW-TECH INDUSTRIES
HYELIN CHOI
Abstract. Although economists and policy makers have devoted considerable atten-
tion to the technology spillovers from foreign to domestic firms, they have not come
to a common conclusion. In order to have a deep understanding of the complex FDI
knowledge spillovers, this study disaggregates the total spillovers into high- and low-
tech industries, in contrast with earlier work that have only examined its aggregate
spillovers. I develop a simple theory to explain the mechanism through which new
technology of the foreign firms is transmitted to the domestic firms by analyzing the
endogenous decision of multinational firms on the location of production of intermedi-
ate goods. The model shows different patterns of knowledge spillovers in the high- and
low-tech industries: immediate catch-up to the foreign firm’s advanced technology but
unsustainable technology spillovers in the low-tech sectors while slow catch-up to the
foreign firm’s technology but continual knowledge spillovers in the high-tech sectors.
The U.S. data for the years 1987-1995, broken down into high- and low-tech industries,
support the model. The pattern of knowledge spillovers in the high- and low-tech in-
dustries are hump-shaped, but in low-tech industries it reaches its peak after a sharp
increase while in the high-tech industries it hits its peak following a smooth increase.
Keywords: Multinational corporations, foreign direct investment(FDI), knowledge spillovers,
technology transfer, high-tech industries, low-tech industries
JEL Classification: F23, D92, D83, L24, O14
I am greatly indebted to Paul Evans for his thoughtful guidance. I am also grateful to Pok-Sang Lam, Paulina Restrepo-Echavarria, Deaho Kim, Byoung Hoon Seok, Semin Kim, and seminarparticipants at the Ohio State University Macro Workshop for their helpful comments and suggestions.All errors are my own responsibility.Department of International Macroeconomics and Finance, Korea Institute for International EconomicPolicy, 246 Yangjaedaero, Seocho-gu, Seoul 137-747, Korea; hlchoi@kiep.go.kr.
1
1. Introduction
Differences in income across countries are largely explained by productivity variations,
and technology plays an important role in determining productivity. However, most of
the world’s technology development is conducted in only a few developed countries even
though R&D expenditures in some developing countries, such as China and India, are
recently increasing. Therefore, international diffusion of technology is very important in
reducing a productivity gap and furthermore income variations across countries.
Multinational firms, enterprises that control and manage production establishments
which span two or more countries, have played a fundamental role in transferring capital,
high skilled labor, technology, and final and intermediate products. Many countries
provide the multinational firms with various incentives, such as lower income taxes, tax
holidays, import duty exemptions and subsidies, to attract multinational firms to their
country based on the belief that they generate positive externalities in the host country.
Among the positive externalities, technology transfer and then productivity growth of
the domestically-owned firms are expected most by policy makers.
In fact, there are several plausible arguments that domestic firms obtain benefits from
the entry of the multinational firms, in particular, in the form of technology transfer.
First, the multinational firms are well known to be larger and more productive and to do
more R&D than purely domestically-owned firms. In the year 2008, total R&D spending
by the eight largest multinational firms was larger than that by all individual countries,
except for the United States and Japan(OECD 2010). Also, there are several mecha-
nisms through which modern technology of the multinational firms may be transferred
to the domestic firms: hiring employees away from multinational firms, imitating their
products, or establishing vertical backward/forward linkages with foreign firms. Addi-
tionally, the entry of multinational firms may lead to severe competition in the host
country market and force domestic firms to improve their efficiency. However, since the
multinational firms cannot fully internalize all the benefits of the domestic firms, the
knowledge spillovers may arise improving the host firms’ productive efficiency. Despite
these favorable reasons for technology transfer by multinational firms, empirical evidence
has not yet produced a common conclusion.
Case studies on technology transfer from multinational to domestic firms present mixed
results. Mauritius and Bangladesh experienced a large increase in textile exports follow-
ing the entries of multinational firms(Rhee and Belot, 1989). Considering that exporters
are more productive than non-exporters, it shows evidence of technology transfer by
multinational firms. On the other hand, Germidis(1977) found no evidence of technol-
ogy transfer in his study of 65 foreign affiliates in 12 developing countries.
Also, although many empirical researchers have tried to examine the knowledge spillovers
from foreign direct investment(FDI) over the past few decades, the results are inconclu-
sive. According to the Gorg and Greenway(2004) which reviewed more than 40 empirical
studies on the FDI spillovers, 20 found positive spillovers, 17 cases yielded insignificant
results, and 8 negative knowledge spillovers. Of course, they cover different countries
and time periods, and also apply different econometric and measurement methods. Re-
searchers motivated by these mixed empirical results have attempted to disentangle com-
plicated patterns of FDI spillovers, by figuring out channels of the knowledge transfer and
various factors which influence knowledge transfer process such as absorptive capacity
of the local firms or multinational firms’ nationalities.
However, previous studies have only examined total effects of multinational firms on
the productivity of domestically-owned firms, overlooking that multinational firms may
differently behave in high- and low-tech industries. This paper investigates whether the
knowledge spillovers depend on the technological complexity of industries, and finds that
the spillovers have different patterns in high- and low-tech industries. I first develop a
simple theory which explains a mechanism by which new technology of foreign firms is
transmitted to host country firms by analyzing the endogenous decision of multinational
firms on the location of production of intermediate goods. The main idea behind the
model comes from the two facts: intermediate goods vary in technological complexity
from simple to complex intermediate goods, and the production of complex intermediate
goods in the host country incurs significant loss of efficiency, which is called technol-
ogy transfer cost. Those facts imply that multinational firms in high-tech industries
which need more complex intermediate goods bring most of intermediate goods from
their parent country to avoid high technology transfer cost, building a limited relation-
ship with local firms and restricted channels of technology transfer. On the other hand,
multinational firms in low-tech industries, which mostly use simple intermediate goods,
procure the majority of intermediate goods within the host country, establishing signifi-
cant linkages with local suppliers and large channels of technology transfer. However, as
technology transfer cost decreases, the production of complex intermediate goods would
move to the host country, which is followed by extension of spillover channels and addi-
tional technology transfer. In sum, the model results in different patterns of knowledge
spillovers in high- and low-tech industries: immediate large but unsustainable increases
in spillovers in low-tech sectors, while slow but continual knowledge spillovers occur in
high-tech sectors.
The U.S. firm-level data from Compustat for the years 1987-1995 support the model.
The empirical results show that the patterns of knowledge spillovers in high- and low-
tech industries are hump-shaped, but in the low-tech industries domestic firms rapidly
keep up with the foreign firm’s productive efficiency with no sustainable increases in
spillovers, while in the high-tech industries they reach the foreign firm’s level at very
low speed but there are continual knowledge spillovers. In this empirical study, I adopt
a quadratic regression equation to capture a non-linear relationship between domestic
firm’s performance and technology transfer cost. Second, I use industry-level data on im-
ports shipped to foreign affiliates from their parent companies obtained from the OECD
in order to measure the technology transfer cost. It is based on the assumption that a low
technology transfer cost is associated with a greater procurement of intermediate goods
within the host country, and thus less intra-firm imports of intermediate goods from
parent companies. Third, I apply the semiparametric Olley-Pakes method to properly
measure total factor productivity of domestic firms.
My study is unique in several dimensions. First, I look at knowledge spillovers in
high- and low-tech industries, respectively, since their different technological features
may cause differences in the knowledge spillovers. It challenges the view that all in-
ward FDI is equally valuable in terms of productivity benefits regardless of industries.
However, my results suggest that FDI spillovers depend on technological complexity of
the industries, and also on the technology transfer cost. Second, as briefly mentioned
above, the empirical analysis is based on the data of intra-firm imports as a proxy for
the extent to which multinational firms have interactions with local firms. The previous
studies usually use employment or production share of foreign firms to measure the for-
eign presence in the host country, leaving channels of knowledge spillovers as a black box.
However, this paper suggests explicit mechanism of knowledge spillovers and measure it
with observable data. Third, I show that the multinational firm’s impact on the local
firm’s productivity is better explained with a quadratic function. The existing literature,
excepting for a few studies, is confined to examining a linear relationship between foreign
presence and domestic firms’ productivity. The notable exception is Buckley, Clegg, and
Wang(2007), in which they found that the impacts of the multinational firms on the
domestic firm’s performance have a curvilinear relationship, and also they argue that
the foreign firm’s impacts depend on the nationality of ownership of foreign investors.
The remainder of this paper is constructed as follows: The following section lays out
a model which explains a knowledge spillovers mechanism. Section three describes the
data and estimation strategy. The estimation results are presented in section four, and
section five provides some concluding remarks.
2. Literature review
FDI has been considered as one of the important knowledge spillover channels. Some
papers theoretically support the appearance of multinational firms in the world econ-
omy.(Markusen 1984, Markusen and Venables 1998). Since firm-specific activities such
as R&D, advertising, marketing, and management services have a characteristic of public
goods, multinational firms do not need to duplicate them whenever they open new affil-
iates. Hence, they take advantage of the economies of multi-plant operation and market
expansion by establishing new affiliates in other countries. Also, Rodriguez-Clare(2007)
argues that the gains from openness, which includes not only trade but also other venues,
are much higher than the gains from only trade. In other words, there is another channel
through which countries interact and large positive impacts accrue to the economy. FDI
could be one of the potential channels.
There are a number of case studies on the FDI spillovers. These studies bear mixed
evidence on the role of multinational firms in generating technology diffusion to domestic
firms, some find evidence of the knowledge spillovers but others do not.
Many international economists have attempted to go beyond qualitative case studies.
Following Caves(1974), Globerman(1979a), and Blomstrom and Persson(1983), many
empirical studies have poured out. Even though a large body of literature has tried to
show empirically the existence and degree of the horizontal knowledge spillovers through
multinational firms, they have come to mixed results of strongly positive, negative, or
even insignificant horizontal knowledge spillovers. Keller and Yeaple(2009) and Blom-
strom and Wolff(1994) showed that the presence of foreign firms is associated with the
growth of host firm’s productivity. On the other hand, some papers argued that for-
eign presence does not seem to have positive impact on the local productivity. Aitken
and Harrison(1999) and Blomstrom(1986a) found that foreign presence rather reduces
productivity of domestically-owned firms by the so-called market-stealing effect. Also,
Haddad and Harrison(1993) concluded that foreign firms do not generate knowledge
spillovers in the host country.
The empirical ambiguity has become a major motivation for further studies. Based
on the fact that the previous studies have treated the mechanisms by which the knowl-
edge spillovers occur as a black box, the following papers have tried to explicitly explain
spillover channels. First, Javorcik(2004), and Lin and Saggi(2007) took into account ver-
tical linkages as one of the spillover channels. When multinational firms make contracts
with local firms to purchase intermediate goods, foreign firms may teach local suppliers
how to efficiently produce goods or how to improve the production management to meet
their higher standards of product quality or on-time delivery. Second, worker turnover is
considered as a spillover channel in some papers. The foreign firms provide local workers
with on-the-job training or better work experience. If the workers are hired by domestic
firms when they leave the foreign firms, they would use their advanced techniques and
knowledge for domestic firms. Markusen and Trofimenko(2007), Gorg and Strobl(2005),
and Poole(2007) supported the argument using firm- or establishment-level data. The
third channel is called demonstration effects. The local firms learn a modern technology
of the foreign firms through the imitation or reverse-engineering. Cheung and Lin(2004)
and Hale and Long(2006) showed positive spillovers effects through demonstration ef-
fects, but this channel does not seem to be as strong as previous channels.
Another attempt to resolve the inconclusiveness of the FDI spillovers is to examine
factors that influence the spillovers, such as nationality of the foreign firms, absorp-
tive capacity of the domestic firms, or the characteristics of the foreign firm’s activities.
Griffith, Redding, and Simpson(2003) argued that the further is the distance from the
technology frontier, the greater is the speed of technology transfer, and foreign multi-
nationals play an important role in the technology transfer by pushing the technology
frontier out and so increasing the speed of convergence to the advanced technology. Gorg,
Hijzen, and Murakozy(2009) presented that labor-intensive activities of foreign affiliates
are unlikely to generate positive productivity spillovers while capital-intensive foreign
affiliates increases the productivity of the local firms. Bukley, Clegg, and Wang(2007)
showed that, among multinational firms which open their affiliates in China, the firms
from Hong Kong, Macau, and Taiwan do not generate knowledge spillovers to Chinese
firms while the firms from U.S., Europe, and Japan bring positive externalities to China.
This paper looks at knowledge spillovers in the high- and low-tech industries based
on the idea that different technological complexity of the industries would force multi-
national firms to make a different decision. Also, this paper suggests explicit mechanism
by which new technology of foreign firms are transferred to the local firms within the
same industry by combining vertical and horizontal knowledge spillovers.
3. Theory
3.1. Overview. I employ a model of Keller(2009) to illustrate technology transfer cost
and the endogenous decision of multinational firms on the location of the production
of intermediate goods. Before explaining the details of the framework, I first give a
background of the model; in particular, the mechanism of knowledge spillovers from
multinational to locally-owned firms and the intuition of how they are differently oper-
ating in high- and low-tech industries.
In this world, there are two countries, a parent country and a host country, and two
sectors, high-tech and low-tech sectors. Also, there are four different kinds of agents:
headquarters and affiliates of multinational firms, local suppliers, and domestic firms.
The headquarters of the multinational firms are located in the parent country and expand
the scope of their operations by establishing foreign affiliates abroad. They own advanced
technology (Ramondo(2009)) and provide their affiliates with intermediate goods needed
for the production of final goods. The affiliates set up in the host country produce
final goods by assembling intermediate goods and support host country’s demand. The
other two agents, local suppliers and domestic firms are host country firms, and the
Figure1. Structure of the Model
local suppliers are in upstream and the domestic firms are in downstream industries,
respectively. The local suppliers produce intermediate goods and provide them to several
firms including both foreign affiliates and domestic firms in the downstream industry.
Lastly, the domestic firms are in the same industry as the multinational firms and produce
final goods by sourcing intermediate goods from local suppliers or for themselves. The
structure of the model is described in figure 1.
The final goods, produced by foreign affiliates in the host country, are completed
from an assembly of a range of intermediate goods. Each of the intermediate goods can
either be sourced from parent firms by being shipped from the home country or be
provided by local suppliers within the host country. In other words, some intermediate
goods are produced in the parent country and the other intermediate goods are
produced in the host country, and they are combined into one final good. When the
intermediate good is produced in the host country, the technology needed to produce
the intermediate good has to be transferred to its producers in the host country from
the multinational firms. The technology transfer necessarily involves some errors
because of incomplete communication between them. It makes host production less
efficient than home production, and the loss of efficiency in the production of
intermediate goods is called technology transfer cost. On the other hand, when the
intermediate goods are produced in the parent country, there is no loss of efficiency in
the intermediate goods production, but the goods have to be delivered to the affiliates
abroad, incurring a shipping cost. Therefore, the multinational firms have to decide the
location of production of intermediate goods by comparing the technology transfer cost
and the shipping cost.
The intermediate goods vary in technological complexity, from simple to complex
intermediate goods. Complex intermediate goods are less likely to be completely
described in the manual or in the face-to-face communication. Even a very detailed
manual cannot cover all the cases of various production environments of the complicate
intermediate goods. Consequently, a large loss of efficiency in the complex goods
production process accrues, requiring higher technology transfer costs than shipping
costs. Therefore, multinational firms minimizing production cost decide to produce
complex intermediate goods at home and then export those goods to their affiliates
abroad. On the other hand, technologically simple intermediate goods generate small
errors in transferring related knowledge, resulting in lower technology transfer costs
than trade costs. Hence, the simple intermediate goods are sourced within the host
country, incurring technology transfer costs.
Now, let’s think about the mechanism of the knowledge spillovers and look at how
the multinational firm’s decision on the intermediate goods works in the mechanism.
I provide an explicit mechanism of the horizontal knowledge spillovers by combining
vertical spillovers which is already strongly supported in the previous literature.
As mentioned previously, since the multinational firms are large in size and have
advanced technology, their entry to the host country may accompany technology diffu-
sion to the domestic firms through various channels, such as worker turnover, product
imitation, or vertical linkages. Among them, the knowledge spillovers through verti-
cal backward linkages have been proved by several papers such as Javorcik(2006) and
Kugler(2006). Also, anecdotal evidence introduced in Javorcik(2004b) confirms verti-
cal knowledge spillovers. When a Czech firm producing aluminum alloy castings and
supplying them to the automotive industry made a contract with a multinational firm,
technicians of the multinational firm regularly visited the Czech firm to teach them how
to improve quality control. Afterward, the Czech firm applied these improvement to
its other production management. Indeed, the multinational firms may directly transfer
their new technology to the local suppliers to meet their high standards for the product
quality or to source intermediate goods at lower prices.
With the vertical knowledge spillovers and the fact that local suppliers usually make
contracts with several firms in the downstream industry including both foreign affiliates
and domestic firms and provide intermediate goods to them, we can infer that advanced
technology of the multinational firms are delivered first to the local suppliers and then
to the domestic firms in the same industry with the foreign affiliates. Namely, the high
quality intermediate goods produced by local suppliers according to the multinational
firm’s new technology are provided to the foreign affiliates as well as domestic firms,
and thus the domestic firms can take advantage of the more efficient intermediate goods
from the entry of foreign affiliates. The intended diffusion of the knowledge between
multinational firms and local firms on the first stage of the mechanism is called technology
transfer, and the inadvertent diffusion in the second step of the mechanism is called
knowledge spillovers. This is graphically shown in figure 2.
Combining the decision of the multinational firms on the production location of the
intermediate goods and the mechanism of the knowledge spillovers leads to different
implications in high- and low-tech sectors. First, since high-tech sectors use more tech-
nologically complex intermediate goods to produce their final goods, and the complex in-
termediate goods generate expensive technology transfer costs, multinational firms decide
to bring a large portion of intermediate goods from the home country to avoid the high
transfer cost. Considering the mechanism of the knowledge spillovers, fewer contracts are
Figure 2. Mechanism of the Technology Transfer
made between multinational firms and local suppliers in the high-tech industries, leading
to less vertical knowledge spillovers, to less intermediate goods produced according to
the foreign firm’s modern technology, and to less horizontal knowledge spillovers. On the
contrary, since low-tech sectors mostly use simple intermediate goods which have lower
technology transfer costs, multinational firms purchase more intermediates goods within
the host country paying the technology transfer cost. It implies large knowledge transfers
from the multinational to the local suppliers and more adoption of advanced intermediate
goods by domestic firms. To summarize, knowledge diffusion through multinational firms
is more active in the low-tech industry due to the cheaper technology transfer cost and
more connection between foreign clients and the domestic suppliers. On the contrary,
the knowledge spillovers in the high-tech industry are restricted because of the expensive
technology transfer cost and less interaction between foreign and domestic firms.
Thus far, we have discussed knowledge spillovers at a fixed technology transfer cost.
Now, let’s consider the knowledge spillovers in a dynamic of the cost. As the technology
transfer cost decreases due to some changes in the global environment, it has different
impacts on the high- and low-tech industries. First, high-tech sectors are largely affected
by the decline in the technology transfer cost because they have largely relied on home
production. The home production of the complex intermediate goods can be moved
to the host country, followed by more contracts between foreign firms and the local
suppliers and continuous knowledge transfer from the foreign to the host firms. On the
contrary, low-tech sectors are not significantly affected by the decline in the technology
transfer cost because most of their technologically simple intermediate goods are already
produced in the host country and they do not have much complex intermediate goods
which benefit from the change in the technology transfer cost. As a result, no additional
backward linkages are created and also no further knowledge spillovers occur. In sum, the
knowledge spillovers are expected to be slow but continuous in the high-tech industries,
but rapid and unsustainable in the low-tech industries.
3.2. The Model. This section provides more details on this framework. Consider a
world which consists of two countries, parent country and host country. Each country
is endowed with L units of labor, which is one of the input factors in the production of
goods. In each country, the representative consumer has homothetic preferences over a
variety of goods produced in the high- and low-tech sectors and a single homogeneous
good Y as the following:
U = Φh ln(´
ω∈Ωhqh(ω)
σh−1σh dω)
σhσh−1 +Φlln(
´ω∈Ωl
ql(ω)σl−1
σl dω)σl
σl−1 +(1−Φh−Φl)lnY · · · (1)
where Φh and Φl are the shares of expenditures spent for goods produced in high-
and low-tech sectors, Ωh and Ωl are sets of varieties available in the respective sectors,
qh(ω) and ql(ω) are the quantities of the output of the variety ω consumed, σh > 1 and
σl > 1 are the elasticities of substitution across varieties in high- and low-tech sectors,
and Y is the quantity of the homogeneous good consumed. Each country produces a
homogeneous good using a single unit of labor, which means wages in both countries are
the same and can be normalized to unity. Assuming that firms are too small to affect
industry level demands, equation (1) implies the following iso-elastic demand function:
qh(ω) = Φh
Ph(ph(ω)
Ph)−σh , ql(ω) = Φl
Pl(pl(ω)
Pl)−σl · · · (2)
where ph(ω) and ph(ω) are the prices of the variety ω in high- and low-tech industries,
and Ph and Pl are the aggregate prices defined as Ph ≡´
ω∈Ωhph(ω)1−σhdω and
Pl ≡´
ω∈Ωlpl(ω)1−σldω in respective sectors.
Each firm produces a different variety of differentiated goods by assembling a contin-
uum of intermediate inputs according to the following production function:
xi = AiLαi [´∞
0 βimi(z)dz]1−α · · · (3)
where z is an index of technological complexity of an intermediate good, mi(z) is the
quantity of an intermediate z used in the production, and βi is a stock of knowledge of
the intermediate good. I will distinguish βDi and βF
i as the domestic and foreign stock
of knowledge later in the paper.
Let’s assume that intermediate goods composition function, mi(z), has the following
functional form:
mi(z) = ϕiexp(−ϕiz) · · · (4)
The firms which use more technologically complex intermediate goods to produce a
final good have higher value of mi(z) for higher z, classifying them as high-tech industries.
On the other hand, low-tech industries use less technologically complex intermediate
goods in producing their final goods and it comes with lower mi(z) for higher z. The
advantage of this functional form is that an average technological complexity of firm i
can be summarized by one parameter, as an inverse of ϕi. That is, firms in the high-tech
industry have low values of ϕi while low-tech firms have high values of ϕi.
The multinational firms open their affiliates abroad and the foreign affiliates produce
final goods according to the production function above. The headquarters of the multi-
national firms must make a decision from where each intermediate good is sourced to
their affiliates. Each intermediate good can be provided either by parent firms from the
home country or by local suppliers within the host country, accompanying trade costs
or technology transfer costs, respectively.
In the case that the intermediate goods are produced by headquarters in the parent
country, no loss of efficiency is incurred in the production of intermediate goods but
shipping costs to deliver the intermediate goods internationally are incurred. The ship-
ping cost is assumed to take the iceberg form, which is widely used in the international
economics literature: τ > 1 units must be shipped in order for one unit to arrive at the
host country. The parameter τ is assumed to be constant regardless of the intermediate
good.
On the other hand, the latter case, the procurement of intermediate goods from local
suppliers, incurs some loss of efficiency in the production of intermediate goods instead of
trade cost because there may be some errors in the communication between headquarters
and local firms. Let ˜λ(t) be the probability of successful communication between head-
quarters and local suppliers. It naturally takes values between 0 and 1. The probability
is independent of the intermediate goods but it depends on the circumstances in which
the international communication takes place. The condition describes how easily and
correctly the communication between headquarters and local suppliers is taking place.
It indicates all economic and social barriers of domestic sourcing involving availability
of high-skilled workers in the host country, geographic distance between home and host
country, or development of the skill describing a blueprint of the intermediate good. It
is simply summarized by a single parameter of t and it is applied for all intermediate
goods. It varies over time. For instance, as the number of high-skilled workers in the
host country increases, local firms may become more efficient at absorbing advanced
technology delivered by foreign firms and thus decrease possible errors in the production
of intermediate goods. Also, as airfares decrease, it may increase the chance of face-
to-face communication, eliminate distorted communication between headquarters and
local suppliers, and so the technology transfer cost may decrease. The change of the
communication circumstances affects all intermediate goods.
Based on the assumption for general probability of successful communication, each
specific intermediate good’s perfect communication probability can be expressed as ˜λ(t)z.
Since ˜λ(t) is between 0 and 1, the higher z of a more complex intermediate good implies
more failure for the successful communication of the required knowledge. It also can
be expressed with a large number of labor units needed to produce the goods as the
following after some simple algebra:
1(λ(t))z
= exp(−zln ˜(λ(t)) = exp(λ(t)z) where λ(t) ≡ −ln( ˜λ(t)) · · · (5)
It is called the technology transfer cost for the intermediate good z. The technology
transfer cost for a specific intermediate good depends both on the circumstances in which
the communication is taking place and the technological complexity of the intermediate
good. For instance, the technology transfer cost of the complex intermediate goods in
unfavorable communication circumstances is very high.
The cost-minimizing multinational firms determine the location of the production
of the intermediate goods by comparing trade cost and technology transfer cost. The
marginal cost of the intermediate good depends on the location where it is produced as
given by:
c(z) =
τ if imported from parent firms
exp(λ(t)z) if produced by local suppliers
· · · (6)
The trade cost is not intermediate good specific, but the technology transfer cost
depends on the technological complexity of the intermediate good, increasing with the
knowledge intensity of the intermediate good. From the equation (6), we can obtain
a cutoff intermediate input, ˆz(t), at which trade cost and technology transfer cost are
equated as given by:
ˆz(t) = 1λ(t) ln(τ) · · · (7)
All intermediate goods with z less than the cutoff level ˆz(t) have lower technology
transfer costs than shipping costs and are thus sourced to the affiliates within the host
country, while all intermediate goods with z higher than ˆz(t) incur higher technology
transfer cost than trade cost and thus are delivered to the affiliates from the parent
country.
Combining the intermediate goods composition function, m(z), and the cutoff level of
the technological complexity of the intermediate inputs, ˆz(t), makes it possible to obtain
the total quantity of intermediate goods procured in the host country by integrating the
intermediate input function up to ˆz(t):
H = 1 − exp(−ϕiˆz(t)) · · · (8)
This is shown in the figure 3 as the area under the exponential function up to z.
Since an average complexity of the intermediate goods is higher in the high-tech in-
dustries than that in the low-tech industries, as mentioned previously, it is summarized
in a single parameter as a lower ϕ for the high-tech and a higher ϕ for the low-tech
industries. Therefore, in the figure above, the high-tech industry is represented by a
smoothly decreasing curve while the low-tech industry by a curve with a steep slope. In
terms of the host country sourcing of the intermediate goods, it is larger for the low-tech
industries because host production is cheaper than home production for simple goods,
whereas local purchase of high-tech foreign firms is smaller to avoid expensive technology
transfer costs. This is supported by data. The figure 4 shows the share of intermediate
Figure 3. Intermediate Goods Composition Function for High and Low-tech industries
goods shipped to the affiliates from their parent company in total sales in the automobile
and food industries, which represent high- and low-tech sectors, respectively. The auto-
mobile industry brings more intermediate goods from the parent country, implying less
sourcing within the host country, while food industries import less intermediate goods,
implying more purchase from the local suppliers.
Now, let’s turn to the mechanism through which a foreign firm’s modern technology
is diffused in the host country. In order to explicitly describe the extent to which foreign
advanced technology have impacts on the domestic firm’s productivity, I compare the
domestic firm’s total output in the absence and presence of the multinational firms in
the host country. Before proceeding to the comparison, let’s recall the stock of knowl-
edge for intermediate goods. As previously mentioned, I would distinguish domestic
and foreign knowledge stocks for intermediate goods. For the simplicity, it is assumed
that the domestic and foreign knowledge stocks are constant for all intermediate goods,
implying that domestic and foreign stocks of knowledge differs by the same amount for
all intermediate inputs. Also, the foreign knowledge stock is assumed to be higher than
the domestic knowledge stock, which is consistent with the fact that multinational firms
are generally large and own advanced technology. The higher stock of knowledge means
Figure 4. Intra-Firm Imports Shares in Automobile and Food Industries
that the intermediate good is more efficiently used in the production process and it will
increase the quantity of output produced. The assumptions can be summarized as the
following:
βF = γ, βD = δ, βF − βD > 0 for all z · · · (9)
First, if there is no presence of the multinational firms in the host country, the domestic
firms produce their final goods by assembling intermediate goods produced relying fully
on the domestic knowledge stock:
YD = ALα[´∞
0 βDmi(z)dz]1−α · · · (10)
where YD is domestic firm’s output, βD is domestic knowledge stock of the intermediate
goods, and the rest of the parameters are the same as those introduced in the
production function.
Second, if multinational firms enter the host country, open their affiliates and make
contracts with local suppliers to purchase intermediate goods, technology transfer
arises through backward linkages, local suppliers produce intermediate goods with
better quality, and domestic firms have access to the high quality intermediate goods.
However, it is only applied for the intermediate goods which have lower technology
transfer costs than trade costs and thus sourced to the affiliates within the host
country. The domestic firm’s output in the presence of the multinational firms is
expressed as the following:
YD = ALα[´ z
0 βF mi(z)dz +´∞
zβDmi(z)dz]1−α · · · (11)
where βF represents foreign knowledge stock. The domestic firms produce final goods
by assembling new simple intermediate goods with better quality and existing more
complex intermediate goods. Since βF is larger than βD by assumption, output of the
domestic firms in the presence of multinational firms is larger than that in the absence
of them.
I will first consider the positive effect that the domestic firms gain from the multina-
tional firms, called knowledge spillovers, in the case in which the general probability of
communication, λ(t), is fixed, and then investigate it in the case where it decreases due
to the improvement in the communication environment between foreign and local firms.
The knowledge spillovers in the former case can be calculated by subtracting equation
(10) from the equation (11). For the simplicity, let´∞
0 βDmi(z)dz be M . Then, it is as
the following:
YD = ALα[(βF − βD)H + M ]1−α − M1−α · · · (12)
where H comes from the equation (8) and it indicates the total amount of the
intermediate goods purchased by foreign affiliates within the host country. Since Hl, for
the low-tech industry, is larger than Hh, for the high-tech industry, the knowledge
spillovers is larger in the low-tech industry than those in the high-tech industry.
Turning to the dynamic patterns of the knowledge spillovers, let’s consider how they
are changing as technology transfer cost decreases. When the technology transfer cost
decreases, the production of complex intermediate goods moves from the home to the
Figure 5. The Impact of the Reduction in TTC on the Cut-off Level of IntermediateGood and Local Sourcing
host country. In the model, the cut-off level of intermediate goods, ˆz(t), moves to the
right and host production, H, increases, as shown in the equation (7) and (8), and it
is described in the figure 5. Due to the downward movement of the technology transfer
cost, the range of intermediate goods of host sourcing is extended, and total amount of
intermediate goods purchased within the host country rises. It implies additional con-
tracts of the multinational firms with the local firms and accordingly expanded channel
of the technology transfer.
Following the reduction in technology transfer cost, its marginal effect on the
domestic output is mathematically shown as given:
∂(YD)∂z
= ALα(1 − α)[(βF − βD)(1 − exp(−ϕiz)) + M ]−α(βF − βD)ϕiexp(−ϕiz) · · · (13)
There are two important points that we can infer from the equation (13). First, the
marginal increase in domestic output decreases in z. The decrease in the technology
transfer cost at its sufficiently high level has small impact on the local firms’ produc-
tivity. It is entirely because of the decreasing intermediate goods composition function.
Actually, it may be that it is more reasonable to assume that the intermediate goods com-
position function for the high-tech industries is increasing with respect to the complexity
of the intermediate goods. That is, their use of intermediate goods is concentrated on
the complex intermediate goods. Even in this case, since the fact that low-tech industries
use relatively more technologically simple intermediate goods than high-tech industries,
the main conclusion of the model is not affected. Second, more importantly, the equation
(13) allows us to observe different behavior of the knowledge spillovers in each sector. To
lay out these explanations precisely, let’s first compare the equation (13) for high- and
low-tech industries. Let an intersection of the intermediate goods composition function
for high- and low-tech sectors be z*, as shown in the figure 3.
exp(−ϕlz) > exp(−ϕhz) for z < z∗ ⇒ (∂(YD)∂z
)L > (∂(YD)∂z
)H for z < z ∗ · · · (14)
exp(−ϕhz) > exp(−ϕlz) for z > z∗⇒(∂(YD)∂z
)H > (∂(YD)∂z
)L for z > z ∗ · · · (15)
The marginal impacts of the technology transfer cost on the domestic firm’s produc-
tivity are reversed once at the intersection, z*, for low-tech industries from higher to
lower impacts and for high-tech industries from lower to higher impacts.
When the communication between foreign and local firms are not efficient and thus
simple intermediate goods are only sourced from local suppliers, the low-tech industries
which mostly use simple intermediate goods establish considerable connection with local
firms and then large channels of technology transfer. On the other hand, under the
circumstances of the inefficient communication between foreign and local firms, high-tech
industries bring most of the intermediate goods from their parent companies, making a
limited interaction with local suppliers and thus restricted technology transfer channels.
However, as an environment of the technology communication between foreign and local
firms has improved, complex intermediate goods’ production has moved from the parent
to the host country, expanding technology transfer channels, in particular, in high-tech
industries which use more complex goods in their production. To summarize, interpreting
it in terms of the host firms, low-tech sectors achieve the advanced level of multinational
firm’s technology shortly, and there is no longer increases in spillovers ever since the
full achievement of the new technology. On the other hand, the domestic firms in the
high-tech sectors do not immediately experience a vast knowledge transfer from foreign
firms, but they continually benefit from the decline in technology transfer cost and gain
productivity growth.
It is shown in the figure 6 in which the horizontal axis represents a reverse of the
technology transfer cost or the extent to which multinational firms interact with the
local firms, and the vertical axis stands for the productivity of the domestic firms. Here,
multinational firm’s technology is standardized to 1 on the vertical axis.
This section has laid out the model which explains the mechanism of the knowledge
spillovers from multinational firms and its different patterns in the high-tech and low-
tech sectors. The next section look at the data and investigate whether the model is
supported by the data.
Figure 6. The Patterns of Knowledge Spillovers in High- and Low-tech industries
4. Data and Methodology
4.1. Methodology.
Most of the previous literature has tested whether the productivity of the domestic
firm is growing faster in industries in which multinational firms are actively operating,
in the form of a linear relation of domestic firm’s productivity and foreign presence in
the industry to which the domestic firm belongs. This study is markedly contrasted to
them in the way that it looks at how the domestic firm’s productivity is affected by the
change in technology transfer cost or the extent to which multinational firms have
linkages with the local firms, not foreign firm’s share in the same industry.
Furthermore, it is tested in the form of a quadratic equation, which allows for a
possibility of diminishing marginal impacts of the technology transfer cost on the
productivity as laid out in the previous section. The following regression equation is
estimated:
tfpijt = β1Xijt + β2zjt + β3z2jt + εijt · · · (16)
where tfpijt is the productivity of the domestic firm i in industry j at time t, z is the
cut-off level at which technology transfer cost and trade cost of the intermediate good
are equated, X is a set of control variables, and εijt is a mean-zero error term.
The model says that the reduction in the technology transfer costs has different im-
pacts on the productivity of the domestic firms in the high-tech and low-tech industries.
The technology transfer from multinational firms is initially very large in the low-tech
industry, but it does not continue as domestic firms become able to perfectly imitate the
advanced technology of the foreign firms. In contrast, the technology transfer is small
in the high-tech industries but it persistently takes place as the technology transfer cost
decreases. If the data is consistent with the conclusion of the model, the coefficient on
z should be positive in both sectors, but small in the high-tech and large in the low-
tech sectors. Also, the coefficient on z2 should be negative in both sectors, but small in
absolute terms in the high-tech and large in the low-tech sectors.
One of the main interesting variables in this equation is z. However, it is unobserved
and difficult to directly measure from the data. Fortunately, equation (8) can be solved
for z to attain:
z = − 1ϕi
ln(1 − H) · · · (17)
1 − H in equation (17) corresponds to the total intermediate goods imported to the
foreign affiliates from their parent company, a variable that can be observed. As a
result, z can be determined from the data on imports of intermediate goods from the
headquarters to the affiliates.
Also, the domestic firms’ productivity is the focus of this paper and thus the consis-
tent estimation for the estimates is crucial. The productivity of the domestic firms are
carefully measured by the Olley-Pakes method. Since OLS does not take into account
the fact that production inputs may be chosen by firms based on their productivity, it
may involve a simultaneity problem and generate biased estimates. Also, OLS does not
consider entry and exit of firms. The Olley-Pakes method deals with these problems by
Table 1. OLS and Olley-Pakes Input Elasticity Estimates
developing a framework in which firms optimally choose investment and sales and also
make entry and exit decisions, thus resolving these problems. The table 1 presents a
comparison of the estimation results of OLS and Olley-Pakes(OP) method. The smaller
O-P coefficients on labor and materials and a larger O-P coefficient on capital than OLS
estimates, confirm that Olley-Pakes method corrects for the biases.
With Olley-Pakes input elasticity estimates calculated above, we can compute total
factor productivity by the following Cobb-Douglas production function:
tfpit = yit − βOPk kit − βOP
l lit − βOPm mit · · · (18)
where yit, kit, and mit stand for the logarithm of output, labor, and material inputs,
respectively, and βOPk , βOP
l , and βOPm are the Olley-Pakes estimates for the elasticities
of output with respect to capital, labor, and materials.
In order to better isolate the FDI spillovers through the mechanism introduced in the
model, I consider several control variables that may influence local firms’ performance.
First, R&D expenditure and a ratio of capital to labor are included because they may
impact firms’ productivity and foreign firms’ activities in the host country, and thus
may cause biased estimates. Also, an index for industry concentration, the Herfindahl-
Hirschman Index(HHI), are considered as one of the control variables. It is because
that the entry of the foreign firms may increase competition with domestic firms, forcing
domestic firms to improve their efficiency.
4.2. Data. The empirical analysis is based on U.S. manufacturing firms for the
years of 1987 through 1995, taken from Standard and Poor’s Compustat database. It
includes publicly traded firms; more importantly, most large U.S. firms. As a result, it
covers most U.S. economic activity. The sample consists of about 1,000 firms and it is
further reduced by deleting those that have missing values or fail to satisfy some standard
criteria. Compustat provides the firm-level data needed to calculate productivity such
as output, labor, capital, and materials. It also offers data on R&D expenditure. First,
the output is measured by net sales from Compustat, and it is deflated by industry-
level price index from the NBER-CES manufacturing database. Labor is measured by
total working hours which is the number of employees from Compustat multiplied by
industry-level average production working hours from the NBER-CES manufacturing
database. Capital is measured by the value of property, plant and equipment, net of
depreciation from Compustat, and it is deflated by deflators from the BEA satellite
accounts. Lastly, materials follow the definition of cost of goods sold plus administrative
and selling expenses less wage expenditures, where the wage is calculated by multiplying
the number of employees with the average industry wage. The former two variables
come from Compustat, while the latter is obtained from the NBER-CES manufacturing
database. The R&D expenditure and capital-labor ratio, used as control variables in
the empirical estimation, also comes from Compustat. R&D expenditure is obtained by
deflating research and development expenses from Compustat by deflators from the BEA
satellite accounts, and the capital-labor ratio is obtained by dividing capital calculated
above by the number of employees. The other control variable, the Herfindahl-Hirschman
Index(HHI), is obtained from BEA and it is transferred from SIC to ISIC Rev.3 using
SIC and ISIC Rev.3 matching table published by Princeton University.
The main purpose of this paper is to examine whether productivity of the domestic
firm is affected by the change in the technology transfer cost and whether it behaves
differently in high- and low-tech sectors. As explained above, the change in technology
transfer cost is reflected in the cutoff-level of intermediate goods, and it is measured by
equation (17) and data on intra-firm imports. The OECD provides data on the activities
of the multinational firms from a wide range of perspectives in their series of ’Measuring
globalization’. In particular, they provide data on intra-firm imports, value-added, and
sales of the foreign affiliates operating in OECD countries. For the U.S., since the data
on the sales of the foreign affiliates are not available, 1 − H is measured by the ratio of
the intra-firm imports to the value-added and then z is calculated using the equation
(17). Since the OECD data is classified by ISIC Rev.3 and the firm-level data from
Compustat use NAICS, I convert NAICS into ISIC Rev.3 referring to correspondence
table published by the United Nations Statistics Division in order to raise compatibility
between the OECD and the Compustat data.
In order to investigate the main question related to the different patterns of knowl-
edge spillovers in the high- and low-tech industries, the full sample is divided into two
groups, according to the share of R&D expenditures in total sales averaged across firms
in the same industry. The group with high shares of R&D expenditures is classified as
high-tech, and the group with low ratio of R&D expenditures in total sales as low-tech.
The intra-firm import shares are 60.14 and 30.39 percentage for high- and low-tech in-
dustries, respectively. The high-tech industries include chemical products, radio, TV &
communication eq., office and computing machinery, wood products, scientific instru-
ments, non-electrical machinery nec, and motor vehicles, while the low-tech industries
cover food, beverages, tobacco, non-metallic mineral products, paper, printing, and pub-
lishing, textiles, clothing, leather, and footwear, electrical machinery nec,basic metals,
other transport equipment, and other manufacturing, and fabricated metal products.
Table 2. Intra-Firm Imports by Industry, in percent of value added
5. Results
This section presents the results of the paper. First, I show some features of the main
variables, in particular, technology transfer cost. Second, I lay out the main results of
estimating the main empirical estimation equation, provide an interpretation of coef-
ficients, and see whether it is consistent with the model or not. Lastly, I reestimate
variants of the main equation to check the robustness of the results.
The technology transfer cost is measured by equation (17) using data on imports
shipped to foreign affiliates from their parent company. The table 2 lists industries
ranked by shares of imports delivered from their headquarters in total value added. It
ranges from about 11 percent in paper, printing, and publishing to over 210 percent
in office and computing machinery. As expected in the model, low-tech industries such
as food, beverages, or textiles import relatively less from their parent companies. In
contrast, high-tech industries such as motor vehicles or computing industry bring more
inputs from their parent company.
The model shows the positive non-relationship between productivity of the domestic
firms and the extent to which multinational firms are connected to the host firms. Also,
the model predicts different patterns of productivity changes in high-tech and low-tech
industries based on their technological complexity of the intermediate goods. The raw
data shows how they are correlated in the following graph.
The horizontal axis represents cut-off level of intermediate goods and the vertical axis
stands for productivities of the domestic firms. The figure 7 provides an evidence that
the productivities of the domestic firms increase as technology transfer cost decreases,
and they are nonlinearly related, rather than the linear association mostly assumed in
the previous literature.
Before proceeding to see the main results, I examine whether the increasing convex
shape fits the data better than the linear one. In order to do so, I compare the regression
results obtained under the assumption of the linear and curvilinear forms, respectively.
As indicated by the adjusted R-squares at the bottom of the table 3, the curvilinear
functional form fits better the data than the linear one. Also, an F-test in which the
linear specification is treated as a restricted equation of the full model shows that the
quadratic term has some power to explain the association of the technology transfer cost
and domestic firms’ productivity at the 1% level of significance.
Table 4 reports the main results. The dependent variable, total factor productivity
for firm i in sector j at time t measured by Olley-Pakes method, is regressed on the
cut-off level of the intermediate goods in sector j at time t in a quadratic function. The
standard errors are clustered by industry-year combinations because firms in the same
industry j encounter the same intra-firm imports and thus the same cut-off level of the
intermediate goods in a given year.
Figure 7. Domestic Productivity and Technology Transfer Cost
Table 3. The comparison of Linear and Curvilinear Functional Form
The first column in the table 4 shows OLS results for the full sample of firms. The
coefficient on the cut-off level of the intermediate goods is positive and statistically signif-
icant and the coefficient on its square is negative and statistically significant, suggesting
Table 4. Impact of Technology Transfer Cost on the Productivity of the DomesticFirms
that there are productivity gains associated with the decrease in the technology transfer
cost but the marginal gains diminish as the technology transfer costs decrease. The point
estimate for zhat, 0.752, shows that productivity of the domestic firms would increase
by 7.52 percentage points when the cut-off intermediate good moves to the right by 10
percent.
The next two columns present results for the subsamples of firms, divided into the
high- and low-tech industries. For both sectors, the coefficients on the cut-off level of
intermediate goods are positive and statistically significant, and the coefficients on its
square are negative and statistically significant. These results clearly show curvilinearity
in the relationship between the technology transfer cost and domestic firms’ productivity.
However, the magnitude of the coefficients are quite different in the two sectors. The
coefficient on zhat is much larger for low-tech industries, supporting an immediate catch-
up to the foreign firms’ advanced technology in the low-tech industry. On the other hand,
Table 5. Results with Current, One-year Lagged, and Two-year Lagged SpilloversVariables for the Full Sample of Data
the zhat coefficient for high-tech industries is small, implying a slow catch-up to the
foreign firms’ technology. These are consistent with the results expected in the model.
The coefficient on the square of zhat is much larger in an absolute term in the low-tech
industries, providing large diminishing impacts of the technology transfer cost on the do-
mestic productivity in the low-tech industries. Meanwhile, it in the high-tech industries
is very small in an absolute term, pertaining to continuous productivity benefits of the
domestic firms. These estimation results are also consistent with the model in the way
that there are no longer technology transfer beyond some technology transfer cost in the
low-tech industries, whereas high-tech industries continually benefit from the decrease
in the technology transfer cost and thus there are continuous knowledge spillovers.
Table 6. Results with Current, One-year Lagged, and Two-year Lagged SpilloversVariables for High-tech Sectors
Turning to the control variables, coefficients on R&D expenditures appear to be neg-
ative and statistically significant, contrasting to the general expectations for the firms’
R&D activities. The capital-labor ratio estimates are positive but they are statistically
insignificant. Lastly, HHI estimates come in as expected, suggesting that a higher indus-
try concentration compels firms to be more efficient.
In sum, the empirical findings are consistent with the model: an immediate knowledge
spillovers but unsustainable increases in spillovers in the low-tech industries, while very
slow knowledge spillovers but continuous spillovers in the high-tech industries.
The knowledge spillovers from foreign to domestic firms may take time to manifest
themselves. In order to examine it, I reestimate the equation (16) taking one-year and
Table 7. Results with Current, One-year Lagged, and Two-year Lagged SpilloversVariables for Low-tech Sectors
two-year lagged spillovers variables. The table 5 through table 7 present the results for
full, high-tech, and low-tech industries, respectively. For the purpose of comparisons, the
tables include regression results with contemporaneous, one-year lagged , and two-year
lagged spillovers variables in each column.
The coefficients on lagged variables for zhat and squared zhat remain positive and
negative, respectively. With regard to the magnitudes of the coefficients on the lagged
zhat and squared zhat, they do not change much, suggesting that the positive impact of
the technology transfer cost on the domestic firms and diminishing marginal impact of
the knowledge spillovers do not quickly vanish. However, the spillovers appear to be no
longer statistically significant in two periods in the low-tech industries. In order to see
Table 8. Results with a different criterion for high- and low-tech industries
Table 9. Results with different measures of labor inputs
whether the results hold under a different criterion for high- and low-tech industries, I
brake down industries into high- and low-tech industries according to the capital-labor
ratio and reestimate the main equation. The industries with higher capital-labor ratios
are classified as high-tech industries, while the other with lower capital-labor ratios as
low-tech industries. The table 8 presents similar results with the previous main results.
Also, I reestimate the main equation by replacing total working hours with total number
of employees in the calculation of the productivity, and the results are very similar with
the previous one, as shown in table 9. They suggest that the results are robust to the
classification of industries and the measurement of the labor.
6. Conclusion
Even though economists and policy makers have devoted considerable attention to the
effects of multinational firms on the productivity of the domestically-owned firms, they
have not come to a common conclusion. It is due to the fact that knowledge spillovers
by multinational firms in the host country are very complex, depending on various firm,
industry, and country factors. Therefore, we need deeper understanding of FDI spillovers
in various circumstances.
In contrast to the earlier literature, which focused on the total effects of multinational
firms on the domestic firms’ productivity, this paper considers the knowledge spillovers
separately in high- and low-tech industries. Using a simple theory, I show that tech-
nological complexity of industries causes different knowledge spillovers by analyzing the
endogenous decision of multinational firms on the location of production of intermediate
goods. The model is supported by U.S. data for the years 1987-2005.
The model and empirical results coherently show that initial knowledge spillovers in
the low-tech industries are very fast and large but there is no longer productivity gains
once domestic firms catch up with the foreign firms’ productive efficiency level. In con-
trast, knowledge spillovers in the high-tech industries are slow but there are continuous
productivity benefits from multinational firms.
This study puts a step forward in understanding the mechanism of the spillovers
from multinational firms and provides new insight by breaking down industries into
high-tech and low-tech groups according to the average complexity of the intermediate
goods. Also, this study embraces the mixed evidence for spillover effects shown in the
previous literature: an economy with a large share of low-tech multinational firms and
low technology transfer cost would experience small or even negative spillovers, one with
a large share of low-tech multinational firms but high technology transfer cost would
be related to considerable knowledge spillovers, and one with a large share of high-
tech foreign firms and low technology transfer cost would be associated with positive or
insignificant spillovers. Therefore, the impacts of the multinational firms on the host
country firms’ productivities cannot be generalized as positive or negative, and they
depend on industrial structure of foreign firms and technology transfer cost.
My study raises several issues for further research. First, one important question is
whether my results extend to other countries, in particular, developing countries. The
availability of the data on intra-firm imports for developing countries would provide more
plentiful evidences for my model. Second, I assume a constant trade cost in the model,
but it actually varies in trade policies or remoteness of a country. Relaxing the assump-
tion would give more insights in understanding complex knowledge spillovers. Lastly,
although the data shows negative impacts of multinational firms on the host country in
the low-tech industries, my model does not explain it. It may be due to that market
competition or other negative factors dominate positive effects. The identification of
those factors and their inclusion in the model would more strongly support the empirical
results. More research needs to be done to better understand the complex multinational
firms’ knowledge spillovers by considering firm, industry, and country dimensions.
References
[1] Aghion, P., and Peter H., “A Model of Growth Through Creative Destruction.”, Econometrica, Vol
60, pp. 323-51.
[2] Aghion, P., and Peter H., “Market Structure and the Growth Process”, Review of Economic Dy-
namics, Vol. 1, pp. 276-305.
[3] Aitken, B., and Harrison, E. A., (1999), “Do Domestic Firms benefit from Foreign Direct Invest-
ment? Evidence from Venezuela”, American Economic Review, Vol. 89, pp. 605-618
[4] Blomstrom, M., and Persson, H., (1983), “Foreign Investment and Spillover Efficiency in an Under-
developed Economy: Evidence from the Mexican Manufacturing Industry”, World Development,
Vol. 11, pp. 493-501.
[5] Blomstrom, M., and Wolff, E., (1994), “Multinational Corporations and Productivity Convergence
in Mexico”, Oxford University Press.
[6] Buckley, J. P., Clegg, J., and Wang, C., (2007), “Is the Relationship between Inward FDI and
Spillover Effects Linear? An Empirical Examination of the Case of China”, Journal of International
Business Studies, Vol.38, pp. 447-459.
[7] Caves, R.E., (1974), “Multinational Firms, Competition and Productivity in Host Country Mar-
kets”, Economica, Vol.41, pp.176-193.
[8] Charles, I. J., and Hall, J., (1999), “Why do Some Countries Produce So Much More Output per
Worker than Others?”, Quarterly Journal of Economics, Vol. 114, pp. 83-116.
[9] Cheung, K.Y., and Lin, P., (2004), “Spillover Effects of FDI on Innovation in China: Evidence from
the Provincial Data”, China Economic Review, Vol.15, pp. 25-44.
[10] Easterly, W., and Levine, R., (2001), "It´s Not Factor Accumulation: Stylized Facts and Growth
Models", World Bank Economic Review, Vol.15, No.2.
[11] Fosfuri, A., Motta, M., and Ronde, T, (2001), "Foreign Direct Investments and Spillovers Through
Workers’ Mobility", Journal of International Economics, Vol.53, pp. 205-222.
[12] Froot A. K., and Stein, J., (1991), “Exchange Rates and Foreign Direct Investment: An Imperfect
Capital Markets Approach”, Quarterly Journal of Economics, Vol. 106, No. 4, pp. 1191-1217
[13] Globerman, S., (1997), “Foreign Direct Investment and Spillover Efficiency Benefits in Canadian
Manufacturing Industries”, Canadian Journal of Economics, Vol. 12, pp. 42-56
[14] Gorg, H., Hijzen, A., and Murakozy, B., (2009), “The Role of Production Technology for Produc-
tivity Spillovers from Multinationals: Firm-Level Evidence for Hungary”, Kiel Working Papers.
[15] Gorg, H., and Strobl, E., (2001), “Multinational Companies and Productivity Spillovers: A Meta-
Analysis”, Economic Journal, Vol.111, pp.F723-739.
[16] Griffith, R., Redding, S., and Simpson, H., (2003), “Productivity Convergence and Foreign Own-
ership at the Establishment Level”, Centre for Economic Performance.
[17] Grossman, M. G., and Helpman, E., (1991), “Trade, Knowledge Spillovers and Growth”, European
Economic Review, vol. 35, no. 3.
[18] Hadda, M., and Harrison, A., (1993), “Are There Positive Spillovers form Direct Foreign Invest-
ment?: Evidence from Panel Data for Morocco”, Journal of Development Economics, Vol. 42,
pp.51-74.
[19] Hale, G., and Long, C., (2006), “What Determines Technological Spillovers of Foreign Direct In-
vestment? Evidence from China”, Yale Economic Growth Center Discussion Paper 934, New Have,
CT.
[20] Javorcik, B. S., (2004), “Does Foreign Direct Investment Increase the Productivity of Domestic
Firms? In Search of Spillovers through Backward Linkages”, The American Economic Review,
Vol.94 pp.605-627.
[21] Keller, W., and Yeaple, R. S., (2009), “Multinational Enterprises, International Trade, and Pro-
ductivity Growth: Firm-Level Evidence from the United States”, The Review of Economics and
Statistics, Vol.91 pp.821-831.
[22] Lin, P. and Saggi, K., (2005), “Multinational Firms and Backward Linkages: A Critical Survey
and a Simple Model”, In Theodore H. Moran, Edward Graham, and Magnus Blomstrom eds., Does
Foreign Direct Investment Promote Development?, pp. 159-174.
[23] Markusen, R. M., (1984), “Multinationals, Multi-Plant Economies, and the Gains from Trade”,
Journal of International Economics, Vol.16 pp.205-226.
[24] Markusen, R.M., and Trofimenko, N., (2007), “Teaching Locals New Tricks: Foreign Experts as a
Channel of Knowledge Transfer”, NBER Working Paper 12872.
[25] Markusen, R. M., and Venables, J. A., (1998), “Multinational Firms and the New Trade Theory”,
Journal of International Economics, Vol.46 pp.183-203.
[26] Olley, S., and Pakes, A., (1996), “The Dynamics of Productivity in the Telecommunications Equip-
ment Industry”, Econometrica, Vol. 64, pp. 1263-1297.
[27] Prescott, E. (1998), “Needed: A Theory of Total Factor Productivity”, International Economic
Review, Vol. 39, pp. 525–52.
[28] Poole, J., (2007), “Multinational Spillovers through Worker Turnover”, Department of Economics,
University of California, Santa Cruz.
[29] Ramondo, N., (2009), “Foreign Plants and Industry Productivity: Evidence from Chile”, Scandi-
navian Journal of Economics, Vol.111, pp.789-809.
[30] Rodriguez-Clare, A., (1996), “Multinationals, Linkages, and Economic Development”, American
Economic Review, Vol 86 pp.852-873.
[31] Rodriguez-Clare, A., (2007), “Trade, Diffusion, and the Gains from Openness”, NBER working
paper.
[32] Romer, P. M., (1986), “Increasing Returns and Long-Run Growth”, Journal of Political Economy
Vol. 94, pp. 1002 - 1037.
[33] Smeets, R., (2008), “Collecting the Pieces of the FDI Knowledge Spillovers Puzzle”, Oxford Uni-
versity Press, Vol. 23, pp. 107-138.